Data Reading During Hyperparameter Tuning

I try to tune hyperparameters of a neural network that I set up. The plan is to start 5 independent training for each set of hyperparameters. During each training process, I will open 8~16 child processes to read data. Apparently I am opening too many child processes just in order to read the same dataset. I need a way to optimize this setup, particularly data reading. Any advice would be greatly appreciated.